same host / 300 Python files ↗
37.1xfaster clean build
0.056 s radar map versus 2.069 s code-review-graph build.
Why: compact maps, not a full review graph.
Navigate repositories.
Spend less context.
radar builds committed maps and exact source routes. Coding agents read fewer files, spend fewer tokens, and finish navigation sooner.
radar query
$ radar query "token verification"
FINAL SOURCE ANCHOR
src/auth.rs#verify_token
28.74 msfresh query p95
10,000 files
40 / 40top-three
source anchors
60.3 %less uncached input
verified source packets
57.6 %smaller agent skill
368 to 156 tokens
measured advantage
Wins by
doing less.
Radar builds the smallest useful index, returns a source pointer instead of a repository tour, and keeps models out of the lookup path.
faster lookup
2.825 ms radar route versus 321.052 ms code-review-graph search.
Latest query internals: scorer-heavy fresh-query p50 is now 14.65%-25.90% faster; tie-break-heavy route lookup is 79.62% faster.
frozen 40-case ledger / native contracts ↗
15.176xfewer actual tokens
Radar used 919 actual cl100k_base tokens versus TokenSave 13,947; Semble used 86,009, or 93.590x Radar's volume.
Why: exact anchors are smaller than bundled context or semantic snippets.
7 tasks × 3 runs per condition / median sums ↗
63.7%lower reference cost
$0.056409 with radar versus $0.155542 with direct source search; uncached input fell 72.4%.
Why: the agent invokes one exact query instead of exploring files.
agent prompt surfaces / current batch ↗
19accepted token cuts
The generated skill fell 368 → 156 tokens, the navigation contract fell 835 → 462, route-add wrapper keys fell 577 → 152, and a 100-1,000-search daily session projects 159,400-1,594,000 input tokens saved.
Why: Radar budgets the instructions and fallback files agents actually read.
100-1,000 code searches / one session per day ↗
$9.56-$95.64monthly Sonnet 5 input saved
Projection: $0.4402-$4.4020 → $0.1214-$1.2140 per 100-1,000-search daily chat session, or $2.23-$22.32/week.
Why: fewer exploratory source reads become fewer billable input tokens.
Scope matters. The external-tool rows compare different native output contracts on pinned versions; the agent-cost row is a published-rate reference scenario, not a Spark invoice. Broader graph and editing tools can be the better choice when an anchor is not enough.
Fewer files read.
Lower agent cost.
Exact routes replace repository tours. Less context goes in, and the source stays in charge.
Committed maps.
Repository structure stays beside the code.
Verified source.
Exact definition packets reject stale answers and whole-file reads.
Git-aware impact.
Callers, tests, dependencies, and co-change history.
Evidence before complexity.
Abstention looked safe on 34 calibration answers, then returned 4 wrong anchors among 9 holdout directs. Radar published it and rolled it back.
Prompt-budgeted surfaces.
Skills, MAPs, route caches, symbol indexes, MCP schemas, and overview rows are measured as agent input, not treated as free text.
Choose your shape.
Different tools, different jobs. This is where radar is the sharper fit.
| project | shape | signal | radar boundary | radar is better when |
|---|---|---|---|---|
| radar ↗reproduced | maps, lexical index, impact, MCP | 28.74 ms p95 / 40 of 40 top-three | Local, deterministic, source-verifiable navigation. | A source pointer is enough. |
| code-review-graph ↗same-host smoke | AST graph, blast radius, PR review | radar: 37.1x faster build / 113.6x faster fresh p95 lookup | Broader review prediction and language coverage. | You need verified local routes and a lighter index. |
| TokenSave ↗pinned native-contract rerun | semantic graph, tools, edits | 9/40 top-one, 14/40 top-three, 13,947 actual tokens; Radar: 15.176x fewer | Broader context and agent integrations. | You need one bounded source result. |
| Semble ↗pinned native-contract rerun | embeddings plus BM25 search | 12/40 top-one, 15/40 top-three, 18/40 retrieved, 86,009 actual tokens; Radar: 93.590x fewer | Semantic chunk retrieval is its strength. | You need semantic chunk search. |
| codebase-memory-mcp ↗project-reported | knowledge graph / 15 MCP tools | 158 languages claimed | Much broader parser and service surface. | You need a broad language graph. |
| CodeGraph ↗project-reported | semantic graph, vectors, MCP | 32 languages claimed | Broader graph, security, and data-flow work. | You need data-flow or security analysis. |
| Aider repo map ↗documented | tree-sitter map in an AI loop | 1k token default map budget | Continuous chat adaptation is its strength. | You need chat-adaptive context. |
| Repomix ↗different shape | whole-repository context pack | no retrieval comparison | Best when a complete portable snapshot is wanted. | You need the full repository snapshot. |
| RepoWise ↗project-reported | graph, Git history, health | under 30 s / 3,000 files designed | Decision and history intelligence are its strength. | You need history, health, and decisions. |
Rival numbers stay labeled. See the full comparison ↗.